lkml.org 
[lkml]   [2020]   [Jun]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RFC PATCH] ALSA: compress: Fix gapless playback state machine
From
Date
Dne 11. 06. 20 v 11:44 Vinod Koul napsal(a):
> On 11-06-20, 11:09, Jaroslav Kysela wrote:
>> Dne 11. 06. 20 v 10:46 Charles Keepax napsal(a):
>>> On Wed, Jun 10, 2020 at 04:28:20PM +0530, Vinod Koul wrote:
>>>> On 10-06-20, 12:40, Jaroslav Kysela wrote:
>>>>> Dne 10. 06. 20 v 12:07 Srinivas Kandagatla napsal(a):
>>>>>> For gapless playback call to snd_compr_drain_notify() after
>>>>>> partial drain should put the state to SNDRV_PCM_STATE_RUNNING
>>>>>> rather than SNDRV_PCM_STATE_SETUP as the driver is ready to
>>>>>> process the buffers for new track.
>>>>>>
>>>>>> With existing code, if we are playing 3 tracks in gapless, after
>>>>>> partial drain finished on previous track 1 the state is set to
>>>>>> SNDRV_PCM_STATE_SETUP which is then moved to SNDRV_PCM_STATE_PREPARED
>>>>>> after data write. With this state calls to snd_compr_next_track() and
>>>>>> few other calls will fail as they expect the state to be in
>>>>>> SNDRV_PCM_STATE_RUNNING.
>>>>>>
>>>>>> Here is the sequence of events and state transitions:
>>>>>>
>>>>>> 1. set_params (Track 1), state = SNDRV_PCM_STATE_SETUP
>>>>>> 2. set_metadata (Track 1), no state change, state = SNDRV_PCM_STATE_SETUP
>>>>>> 3. fill and trigger start (Track 1), state = SNDRV_PCM_STATE_RUNNING
>>>>>> 4. set_next_track (Track 2), state = SNDRV_PCM_STATE_RUNNING
>>>>>> 5. partial_drain (Track 1), state = SNDRV_PCM_STATE_SETUP
>>>>>> 6 snd_compr_drain_notify (Track 1), state = SNDRV_PCM_STATE_SETUP
>>>>>> 7. fill data (Track 2), state = SNDRV_PCM_STATE_PREPARED
>>>>>> 8. set_metadata (Track 3), no state change, state = SNDRV_PCM_STATE_PREPARED
>>>>>> 9. set_next_track (Track 3), !! FAILURE as state != SNDRV_PCM_STATE_RUNNING
>>>>>
>>>>>
>>>>> The snd_compr_drain_notify() is called only from snd_compr_stop(). Something
>>>>> is missing in this sequence?
>>>>
>>>> It is supposed to be invoked by driver when partial drain is complete..
>>>> both intel and sprd driver are calling this. snd_compr_stop is stop
>>>> while draining case so legit
>>>>
>>>
>>> Not sure I follow this statement, could you elaborate a bit?
>>> snd_compr_stop putting the state to RUNNING seems fundamentally
>>> broken to me, the whole point of snd_compr_stop is to take the
>>> state out of RUNNING.
>>
>> Yes. I agree. It seems that the acknowledge for the partial drain should be
>> handled differently.
>
> Yeah sorry I overlooked that case and was thinking of it being invoked
> from driver!
>
> Yes this would make the snd_compr_stop() behave incorrectly.. so this
> cant be done as proposed.
>
> But we still need to set the draining stream state properly and I am
> thinking below now:
>
> diff --git a/sound/core/compress_offload.c b/sound/core/compress_offload.c
> index 509290f2efa8..9aba851732d7 100644
> --- a/sound/core/compress_offload.c
> +++ b/sound/core/compress_offload.c
> @@ -929,7 +929,9 @@ static int snd_compr_partial_drain(struct snd_compr_stream *stream)
> }
>
> stream->next_track = false;
> - return snd_compress_wait_for_drain(stream);
> + retval = snd_compress_wait_for_drain(stream);
> + stream->runtime->state = SNDRV_PCM_STATE_RUNNING;
> + return retval;
> }

I see a race possibility when the last track is too small and the driver
signals the end-of-track twice. In this case the partial drain should not end
with the running state. It would be probably better to separate partial / last
track acknowledgements.

Jaroslav

--
Jaroslav Kysela <perex@perex.cz>
Linux Sound Maintainer; ALSA Project; Red Hat, Inc.

\
 
 \ /
  Last update: 2020-06-11 12:42    [W:0.087 / U:0.228 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site